Chinese open-weights AI models are winning startup adoption over US closed-model APIs

Short summary

Chinese open-weights AI models like Moonshot's Kimi K3 and Alibaba's offerings are winning startup adoption, with a16z's Martin Casado claiming 80% of startups use Chinese models in production. The key insight is that China treats models as loss leaders to commoditize complements (hardware, ecosystem), while US labs gate models behind APIs. Sticker-price token comparisons are misleading; true cost is tokens-burned-per-task times price-per-token, and Chinese labs compound on Western R&D by distilling frontier outputs without ToS constraints.

• •80% of startups reportedly use Chinese AI models in production
• •China commoditizes models as loss leaders; US labs gate them behind premium APIs
• •True AI cost = tokens burned per task × price per token, not list price per million tokens
• •US policy restrictions on frontier models for security work create self-inflicted competitive wounds

Generated with AI, which can make mistakes.